Knowledge Hub

Responsible Use of AI in Evidence Synthesis

Fifty plain-language answers to the most common questions about using AI responsibly in evidence synthesis — validation, oversight, bias, transparency, reporting and more.

50 questions · 9 topics

Foundations of Responsible AI

6 questions
CKO-001Strong evidence

What is responsible AI in evidence synthesis?

Responsible AI in evidence synthesis is the use of AI systems in ways that preserve research integrity, methodological rigour, transparency, accountability, reproducibility and trustworthiness.

CKO-002Strong evidence

Why is AI being introduced into evidence synthesis?

AI is being introduced because the volume and complexity of research have exceeded what can realistically be managed using fully manual processes.

CKO-003Strong evidence

Who is responsible for AI-assisted evidence synthesis?

The evidence synthesis team remains responsible for all methods, decisions, outputs and conclusions, regardless of which AI tools are used.

CKO-004Strong evidence

Can AI replace evidence synthesists?

No. Current evidence supports AI as a companion to evidence synthesists rather than a replacement.

CKO-008Strong evidence

Is summarising evidence the same as evidence synthesis?

No. Summarisation describes evidence, whereas evidence synthesis systematically evaluates and integrates evidence to support conclusions.

CKO-010Strong evidence

What is the most important principle for responsible AI in evidence synthesis?

AI should only be used when it improves efficiency or capability without compromising rigour, transparency, accountability, reproducibility or trust.

AI Literacy & Tool Adoption

4 questions
CKO-011Strong evidence

What is AI literacy?

AI literacy is the ability to understand, evaluate, use and govern AI systems appropriately and responsibly.

CKO-012Strong evidence

How much technical knowledge do evidence synthesists need?

Evidence synthesists do not need deep technical expertise but should understand enough to evaluate whether a tool is appropriate and trustworthy.

CKO-013Strong evidence

How should I decide whether to use an AI tool?

Use AI only when there is evidence that it is suitable for the task and can be used without compromising methodological rigour.

CKO-014Strong evidence

What evidence should support adoption of an AI tool?

Adoption should be supported by validation studies, independent evaluations, benchmarking results and evidence from similar review contexts.

Validation & Evaluation

7 questions
CKO-005Strong evidence

What is validation?

Validation is the process of determining whether an AI system performs reliably and accurately for its intended task.

CKO-006Strong evidence

What is generalisability?

Generalisability is the extent to which AI performance remains reliable across different settings, topics and review contexts.

CKO-015Strong evidence

What is stability?

Stability is the extent to which an AI system produces consistent outputs when given the same input repeatedly.

CKO-016Strong evidence

What is robustness?

Robustness is the ability of an AI system to maintain acceptable performance when faced with unusual, complex or unexpected inputs.

CKO-017Strong evidence

What is a SWAR?

A SWAR (Study Within A Review) is a structured evaluation embedded inside a real evidence synthesis project.

CKO-037Strong evidence

What is a benchmark dataset?

A benchmark dataset is a standardised dataset used to evaluate and compare AI systems.

CKO-038Strong evidence

What is a cumulative evidence base?

A cumulative evidence base is a collection of comparable studies that collectively improve understanding of AI performance.

AI Across the Review Workflow

5 questions
CKO-018Emerging evidence

Can AI search for evidence?

AI can support evidence searching, but search remains a high-risk task requiring careful validation and oversight.

CKO-019Strong evidence

Can AI screen studies?

Yes. Screening is one of the most promising applications of AI in evidence synthesis.

CKO-020Strong evidence

Can AI extract study data?

Yes. AI can assist with data extraction, particularly when combined with human verification.

CKO-021Emerging evidence

Can AI assess risk of bias?

AI may assist with risk of bias assessment, but human judgement remains essential.

CKO-041Strong evidence

Can AI perform meta-analysis?

Current generative AI systems cannot independently perform a valid meta-analysis.

Human Oversight & Autonomy

4 questions
CKO-022Strong evidence

What is human oversight?

Human oversight is the active involvement of people in supervising, verifying and taking responsibility for AI-assisted work.

CKO-023Strong evidence

What decisions should remain human?

Final methodological, interpretive and accountability decisions should remain human responsibilities.

CKO-024Emerging evidence

What is agentic AI?

Agentic AI refers to systems that autonomously perform multiple linked tasks to pursue a goal.

CKO-025Strong evidence

Are autonomous end-to-end evidence synthesis systems trustworthy?

Current evidence does not support fully autonomous end-to-end evidence synthesis systems for decision-grade evidence production.

Hallucinations, Bias & Fairness

5 questions
CKO-007Strong evidence

What is a hallucination?

A hallucination is information generated by AI that is unsupported, fabricated or incorrect.

CKO-026Strong evidence

What is algorithmic bias?

Algorithmic bias occurs when an AI system produces systematically distorted, unfair or inaccurate outputs due to biases in data, design or implementation.

CKO-027Strong evidence

Can AI amplify bias?

Yes. AI systems can reproduce and sometimes magnify biases already present in research, datasets and human decision-making.

CKO-028Strong evidence

How can bias be reduced?

Bias can be reduced through diverse datasets, transparent evaluation, ongoing monitoring and human oversight.

CKO-044Strong evidence

What is fairness in AI?

Fairness refers to whether AI systems perform appropriately and equitably across different groups and contexts.

Transparency & Reporting

7 questions
CKO-009Strong evidence

What should be reported when using AI?

Researchers should report what AI was used, why it was used, how it was used and how outputs were verified.

CKO-029Strong evidence

What AI use should be reported?

Any AI use that influences review methods, decisions, outputs or conclusions should be reported.

CKO-030Strong evidence

What information should be reported when using AI?

Report the tool, version, purpose, workflow stage, prompts where appropriate and verification procedures.

CKO-031Strong evidence

What is provenance?

Provenance is the documented history of where information originated and how it was processed.

CKO-032Strong evidence

Why is provenance important?

Provenance supports transparency, accountability, auditability and reproducibility.

CKO-039Strong evidence

Why should prompts be documented?

Prompts influence AI behaviour and should therefore be treated as part of the research method.

CKO-040Strong evidence

Why should model versions be documented?

Different model versions may produce different outputs and performance levels.

Ethics, Privacy & Governance

5 questions
CKO-033Strong evidence

What ethical issues arise from AI use?

Key ethical issues include fairness, bias, privacy, transparency, environmental impact and responsible use of data.

CKO-034Strong evidence

What copyright issues arise from AI use?

Researchers must ensure they have appropriate rights to upload, process and share copyrighted materials.

CKO-035Strong evidence

What privacy issues arise from AI use?

Sensitive, confidential or identifiable information may be exposed if AI systems are used inappropriately.

CKO-036Strong evidence

What is responsible handover?

Responsible handover is the process of ensuring users understand an AI tool's purpose, limitations, evidence base and safe use.

CKO-045Strong evidence

What is data governance?

Data governance refers to the policies, processes and controls used to manage data responsibly.

Evidence Quality & the Future

7 questions
CKO-042Strong evidence

What is decision-grade evidence?

Decision-grade evidence is evidence produced using rigorous, transparent and trustworthy methods that support real-world decisions.

CKO-043Strong evidence

Can a chatbot provide decision-grade evidence?

No. Chatbots may support information access but cannot independently produce decision-grade evidence.

CKO-046Strong evidence

Who are the key stakeholders in responsible AI for evidence synthesis?

Evidence synthesists, methodologists, developers, organisations, funders, publishers, trainers and evidence users.

CKO-047Strong evidence

Why should AI evaluations be shared?

Sharing evaluations helps build a cumulative evidence base and improves collective learning.

CKO-048Strong evidence

What is open science in the context of AI?

Open science involves making methods, evaluations, datasets and findings as transparent and reusable as possible.

CKO-049Strong evidence

What is research integrity?

Research integrity refers to conducting research honestly, rigorously, transparently and accountably.

CKO-050Emerging evidence

What is the future of evidence synthesis in an AI-enabled world?

The future is likely to involve human-led, AI-assisted evidence systems that are more efficient, more connected and continuously updated.